Music_genre_classification.ipynb

Kernel: Python 3

Music genre classification notebook

Importing Libraries

# feature extractoring and preprocessing data
import librosa
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
import os
from PIL import Image
import pathlib
import csv

# Preprocessing
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder, StandardScaler

#Keras
import keras

import warnings
warnings.filterwarnings('ignore')

Extracting music and features

Dataset

We use GTZAN genre collection dataset for classification.

The dataset consists of 10 genres i.e

• Blues

• Classical

• Country

• Disco

• Hiphop

• Jazz

• Metal

• Pop

• Reggae

• Rock

Each genre contains 100 songs. Total dataset: 1000 songs

Extracting the Spectrogram for every Audio

cmap = plt.get_cmap('inferno')

plt.figure(figsize=(10,10))
genres = 'blues classical country disco hiphop jazz metal pop reggae rock'.split()
for g in genres:
    pathlib.Path(f'img_data/{g}').mkdir(parents=True, exist_ok=True)     
    for filename in os.listdir(f'./MIR/genres/{g}'):
        songname = f'./MIR/genres/{g}/{filename}'
        y, sr = librosa.load(songname, mono=True, duration=5)
        plt.specgram(y, NFFT=2048, Fs=2, Fc=0, noverlap=128, cmap=cmap, sides='default', mode='default', scale='dB');
        plt.axis('off');
        plt.savefig(f'img_data/{g}/{filename[:-3].replace(".", "")}.png')
        plt.clf()
 

All the audio files get converted into their respective spectrograms .WE can noe easily extract features from them.

Extracting features from Spectrogram

We will extract

• Mel-frequency cepstral coefficients (MFCC)(20 in number)

• Spectral Centroid,

• Zero Crossing Rate

• Chroma Frequencies

• Spectral Roll-off.

header = 'filename chroma_stft rmse spectral_centroid spectral_bandwidth rolloff zero_crossing_rate'
for i in range(1, 21):
    header += f' mfcc{i}'
header += ' label'
header = header.split()

Writing data to csv file

We write the data to a csv file

file = open('data.csv', 'w', newline='')
with file:
    writer = csv.writer(file)
    writer.writerow(header)
genres = 'blues classical country disco hiphop jazz metal pop reggae rock'.split()
for g in genres:
    for filename in os.listdir(f'./MIR/genres/{g}'):
        songname = f'./MIR/genres/{g}/{filename}'
        y, sr = librosa.load(songname, mono=True, duration=30)
        chroma_stft = librosa.feature.chroma_stft(y=y, sr=sr)
        spec_cent = librosa.feature.spectral_centroid(y=y, sr=sr)
        spec_bw = librosa.feature.spectral_bandwidth(y=y, sr=sr)
        rolloff = librosa.feature.spectral_rolloff(y=y, sr=sr)
        zcr = librosa.feature.zero_crossing_rate(y)
        mfcc = librosa.feature.mfcc(y=y, sr=sr)
        to_append = f'{filename} {np.mean(chroma_stft)} {np.mean(rmse)} {np.mean(spec_cent)} {np.mean(spec_bw)} {np.mean(rolloff)} {np.mean(zcr)}'    
        for e in mfcc:
            to_append += f' {np.mean(e)}'
        to_append += f' {g}'
        file = open('data.csv', 'a', newline='')
        with file:
            writer = csv.writer(file)
            writer.writerow(to_append.split())

The data has been extracted into a data.csv file.

Analysing the Data in Pandas

data = pd.read_csv('data.csv')
data.head()

data.shape

(1000, 28)

# Dropping unneccesary columns
data = data.drop(['filename'],axis=1)

Encoding the Labels

genre_list = data.iloc[:, -1]
encoder = LabelEncoder()
y = encoder.fit_transform(genre_list)

Scaling the Feature columns

scaler = StandardScaler()
X = scaler.fit_transform(np.array(data.iloc[:, :-1], dtype = float))

Dividing data into training and Testing set

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

len(y_train)

800

len(y_test)

200

X_train[10]

array([-0.9149113 ,  0.18294103, -1.10587131, -1.3875197 , -1.14640873,
       -0.97232926, -0.29174214,  1.20078936, -0.68458101, -0.55849017,
       -1.27056582, -0.88176926, -0.74844069, -0.40970382,  0.49685952,
       -1.12666045,  0.59501437, -0.39783853,  0.29327275, -0.72916871,
        0.63015786, -0.91149976,  0.7743942 , -0.64790051,  0.42229852,
       -1.01449461])

Classification with Keras

Building our Network

from keras import models
from keras import layers

model = models.Sequential()
model.add(layers.Dense(256, activation='relu', input_shape=(X_train.shape[1],)))

model.add(layers.Dense(128, activation='relu'))

model.add(layers.Dense(64, activation='relu'))

model.add(layers.Dense(10, activation='softmax'))

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

history = model.fit(X_train,
                    y_train,
                    epochs=20,
                    batch_size=128)
                   

Epoch 1/20
800/800 [==============================] - 1s 811us/step - loss: 2.1289 - acc: 0.2400
Epoch 2/20
800/800 [==============================] - 0s 39us/step - loss: 1.7940 - acc: 0.4088
Epoch 3/20
800/800 [==============================] - 0s 37us/step - loss: 1.5437 - acc: 0.4450
Epoch 4/20
800/800 [==============================] - 0s 38us/step - loss: 1.3584 - acc: 0.5413
Epoch 5/20
800/800 [==============================] - 0s 38us/step - loss: 1.2220 - acc: 0.5750
Epoch 6/20
800/800 [==============================] - 0s 41us/step - loss: 1.1187 - acc: 0.6288
Epoch 7/20
800/800 [==============================] - 0s 37us/step - loss: 1.0326 - acc: 0.6550
Epoch 8/20
800/800 [==============================] - 0s 44us/step - loss: 0.9631 - acc: 0.6713
Epoch 9/20
800/800 [==============================] - 0s 47us/step - loss: 0.9143 - acc: 0.6913
Epoch 10/20
800/800 [==============================] - 0s 37us/step - loss: 0.8630 - acc: 0.7125
Epoch 11/20
800/800 [==============================] - 0s 36us/step - loss: 0.8095 - acc: 0.7263
Epoch 12/20
800/800 [==============================] - 0s 37us/step - loss: 0.7728 - acc: 0.7700
Epoch 13/20
800/800 [==============================] - 0s 36us/step - loss: 0.7433 - acc: 0.7563
Epoch 14/20
800/800 [==============================] - 0s 45us/step - loss: 0.7066 - acc: 0.7825
Epoch 15/20
800/800 [==============================] - 0s 43us/step - loss: 0.6718 - acc: 0.7787
Epoch 16/20
800/800 [==============================] - 0s 36us/step - loss: 0.6601 - acc: 0.7913
Epoch 17/20
800/800 [==============================] - 0s 36us/step - loss: 0.6242 - acc: 0.7963
Epoch 18/20
800/800 [==============================] - 0s 44us/step - loss: 0.5994 - acc: 0.8038
Epoch 19/20
800/800 [==============================] - 0s 42us/step - loss: 0.5715 - acc: 0.8125
Epoch 20/20
800/800 [==============================] - 0s 39us/step - loss: 0.5437 - acc: 0.8250

test_loss, test_acc = model.evaluate(X_test,y_test)

200/200 [==============================] - 0s 244us/step

print('test_acc: ',test_acc)

test_acc:  0.68

Tes accuracy is less than training dataa accuracy. This hints at Overfitting

Validating our approach

Let's set apart 200 samples in our training data to use as a validation set:

x_val = X_train[:200]
partial_x_train = X_train[200:]

y_val = y_train[:200]
partial_y_train = y_train[200:]

Now let's train our network for 20 epochs:

model = models.Sequential()
model.add(layers.Dense(512, activation='relu', input_shape=(X_train.shape[1],)))
model.add(layers.Dense(256, activation='relu'))
model.add(layers.Dense(128, activation='relu'))
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10, activation='softmax'))

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.fit(partial_x_train,
          partial_y_train,
          epochs=30,
          batch_size=512,
          validation_data=(x_val, y_val))
results = model.evaluate(X_test, y_test)

Train on 600 samples, validate on 200 samples
Epoch 1/30
600/600 [==============================] - 1s 1ms/step - loss: 2.3074 - acc: 0.0950 - val_loss: 2.1857 - val_acc: 0.2850
Epoch 2/30
600/600 [==============================] - 0s 65us/step - loss: 2.1126 - acc: 0.3783 - val_loss: 2.0936 - val_acc: 0.2400
Epoch 3/30
600/600 [==============================] - 0s 59us/step - loss: 1.9535 - acc: 0.3633 - val_loss: 1.9966 - val_acc: 0.2600
Epoch 4/30
600/600 [==============================] - 0s 58us/step - loss: 1.8082 - acc: 0.3833 - val_loss: 1.8713 - val_acc: 0.3250
Epoch 5/30
600/600 [==============================] - 0s 59us/step - loss: 1.6663 - acc: 0.4083 - val_loss: 1.7302 - val_acc: 0.3450
Epoch 6/30
600/600 [==============================] - 0s 52us/step - loss: 1.5329 - acc: 0.4550 - val_loss: 1.6233 - val_acc: 0.3700
Epoch 7/30
600/600 [==============================] - 0s 62us/step - loss: 1.4236 - acc: 0.4850 - val_loss: 1.5402 - val_acc: 0.3950
Epoch 8/30
600/600 [==============================] - 0s 57us/step - loss: 1.3250 - acc: 0.5117 - val_loss: 1.4655 - val_acc: 0.3800
Epoch 9/30
600/600 [==============================] - 0s 52us/step - loss: 1.2338 - acc: 0.5633 - val_loss: 1.3927 - val_acc: 0.4650
Epoch 10/30
600/600 [==============================] - 0s 61us/step - loss: 1.1577 - acc: 0.5983 - val_loss: 1.3338 - val_acc: 0.5500
Epoch 11/30
600/600 [==============================] - 0s 64us/step - loss: 1.0981 - acc: 0.6317 - val_loss: 1.3111 - val_acc: 0.5550
Epoch 12/30
600/600 [==============================] - 0s 52us/step - loss: 1.0529 - acc: 0.6517 - val_loss: 1.2696 - val_acc: 0.5400
Epoch 13/30
600/600 [==============================] - 0s 52us/step - loss: 0.9994 - acc: 0.6567 - val_loss: 1.2480 - val_acc: 0.5400
Epoch 14/30
600/600 [==============================] - 0s 65us/step - loss: 0.9673 - acc: 0.6633 - val_loss: 1.2384 - val_acc: 0.5700
Epoch 15/30
600/600 [==============================] - 0s 58us/step - loss: 0.9286 - acc: 0.6633 - val_loss: 1.1953 - val_acc: 0.5800
Epoch 16/30
600/600 [==============================] - 0s 59us/step - loss: 0.8849 - acc: 0.6783 - val_loss: 1.2000 - val_acc: 0.5550
Epoch 17/30
600/600 [==============================] - 0s 61us/step - loss: 0.8621 - acc: 0.6850 - val_loss: 1.1743 - val_acc: 0.5850
Epoch 18/30
600/600 [==============================] - 0s 61us/step - loss: 0.8195 - acc: 0.7150 - val_loss: 1.1609 - val_acc: 0.5750
Epoch 19/30
600/600 [==============================] - 0s 62us/step - loss: 0.7976 - acc: 0.7283 - val_loss: 1.1238 - val_acc: 0.6150
Epoch 20/30
600/600 [==============================] - 0s 63us/step - loss: 0.7660 - acc: 0.7650 - val_loss: 1.1604 - val_acc: 0.5850
Epoch 21/30
600/600 [==============================] - 0s 65us/step - loss: 0.7465 - acc: 0.7650 - val_loss: 1.1888 - val_acc: 0.5700
Epoch 22/30
600/600 [==============================] - 0s 65us/step - loss: 0.7099 - acc: 0.7517 - val_loss: 1.1563 - val_acc: 0.6050
Epoch 23/30
600/600 [==============================] - 0s 68us/step - loss: 0.6857 - acc: 0.7683 - val_loss: 1.0900 - val_acc: 0.6200
Epoch 24/30
600/600 [==============================] - 0s 67us/step - loss: 0.6597 - acc: 0.7850 - val_loss: 1.0872 - val_acc: 0.6300
Epoch 25/30
600/600 [==============================] - 0s 67us/step - loss: 0.6377 - acc: 0.7967 - val_loss: 1.1148 - val_acc: 0.6200
Epoch 26/30
600/600 [==============================] - 0s 64us/step - loss: 0.6070 - acc: 0.8200 - val_loss: 1.1397 - val_acc: 0.6150
Epoch 27/30
600/600 [==============================] - 0s 66us/step - loss: 0.5991 - acc: 0.8167 - val_loss: 1.1255 - val_acc: 0.6300
Epoch 28/30
600/600 [==============================] - 0s 62us/step - loss: 0.5656 - acc: 0.8333 - val_loss: 1.0955 - val_acc: 0.6350
Epoch 29/30
600/600 [==============================] - 0s 66us/step - loss: 0.5513 - acc: 0.8300 - val_loss: 1.1030 - val_acc: 0.6050
Epoch 30/30
600/600 [==============================] - 0s 56us/step - loss: 0.5498 - acc: 0.8233 - val_loss: 1.0869 - val_acc: 0.6250
200/200 [==============================] - 0s 65us/step

results

[1.2261371064186095, 0.65]

Predictions on Test Data

predictions = model.predict(X_test)

predictions[0].shape

(10,)

np.sum(predictions[0])

1.0

np.argmax(predictions[0])

8